NATURAL LANGUAGE PROCESSING (NLP) PORTAL FOR THIRD PARTY APPLICATIONS

• US 20140278373A1
• Filed: 03/05/2014
• Published: 09/18/2014
• Est. Priority Date: 03/15/2013
• Status: Abandoned Application

• Current Assignee: Ask Ziggy Inc.
• Original Assignee: Ask Ziggy Inc.
• Inventors: Grove, Kyle Wade, Leib, Shai, Bertran, Thomas, Gandhi, Dipesh
• Application Number:
  US14/198,060
• Publication Number:
  US 20140278373A1
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30

Abstract

A method for generating an natural language processing (NLP) model including obtaining tags, obtaining actions to be implemented by a third party application, obtaining a training corpus including sentences, where at least one word in each of the sentences is associated with one of the tags, and wherein each of the sentences is associated with one of the actions. The method further includes generating features for the NLP model for the third party application using the tags, the actions, and the training corpus, training the NLP model using the features and the training corpus to obtain a trained NLP model, and generating an APIKey for use by the third party application, where the API provides the third party application access to the trained NLP model.

Claims

1. A non-transitory computer readable medium comprising instructions, which when executed perform a method for generating a natural language processing (NLP) model, the method comprising:

obtaining a plurality of tags;

obtaining a plurality of actions to be implemented by a third party application;

obtaining a training corpus comprising a plurality of sentences, wherein at least one word in each of the plurality of sentences is associated with one of the plurality of tags, and wherein each of the plurality of sentences is associated with one of the plurality of actions;

generating a plurality of features for the NLP model for the third party application using the plurality of tags, the plurality of actions, and the training corpus;

training the NLP model using the plurality of features and the training corpus to obtain a trained NLP model; and

generating an APIKey for use by the third party application, wherein the API provides the third party application access to the trained NLP model.

2. The non-transitory computer readable medium of claim 1, the method further comprising:

prior to generating the APIKey;

validating the trained NLP model using a sample sentence, wherein an NLP engine using the trained NLP model tags at least one entity in the sample sentence with one of the plurality of tags and tags the sample sentence with one of the plurality of actions.

3. The non-transitory computer readable medium of claim 2, wherein the trained NLP model is validated when a user confirms that the NLP engine, using the trained NLP model, correctly tagged the sample sentence.

4. The non-transitory computer readable medium of claim 1, wherein the trained NLP model is a maximum entropy Markov model.

5. The non-transitory computer readable medium of claim 1, wherein a tag of the plurality of tags is associated with an entity list comprising entity list items, wherein at least one of the entity list items is used to generate one of the plurality of features for the tag.

6. The non-transitory computer readable medium of claim 1, wherein one of the plurality of features is a word-level feature.

7. The non-transitory computer readable medium of claim 6, wherein the word-level feature is one selected from a group consisting of a feature from a common feature set and a feature for a high frequency word in the training corpus.

8. The non-transitory computer readable medium of claim 1, wherein one of the plurality of features is a sentence-level feature.

9. The non-transitory computer readable medium of claim 8, wherein the sentence-level feature is generated using an informative word and one of the plurality of tags.

10. The non-transitory computer readable medium of claim 9, wherein the informative word is identified from the training corpus using term frequency−

inverse document frequency (tf*idf) scores.

11. The non-transitory computer readable medium of claim 9, wherein the informative word is identified by combining all sentences in the training corpus that are associated with one of the plurality of actions to generate a per-action document, calculating tf*idf scores for each of the words in the per-action document, and identifying the informative word using the tf*idf scores.

12. The non-transitory computer readable medium of claim 8, wherein the sentence-level feature specifies one of the plurality of tags.

13. The non-transitory computer readable medium of claim 8, wherein the sentence-level feature does not specify any of the plurality of tags.

14. The non-transitory computer readable medium of claim 1, wherein the action is associated with a method in the third party application.

15. The non-transitory computer readable medium of claim 1, wherein the APIKey is uniquely associated with the third party application.

16. The non-transitory computer readable medium of claim 15, the method further comprising:

receiving an API call from the third party application, wherein the API call comprises text and the APIKey, wherein the text corresponds to an utterance of a user of the third party application.

17. The non-transitory computer readable medium of claim 16, the method further comprising:

identifying the trained NLP model using the APIKey; and

analyzing the text using the trained NLP model to generate a result, wherein the result specifies one of the plurality of actions.

18. The non-transitory computer readable medium of claim 16, wherein the utterance is an audio utterance.

19. The non-transitory computer readable medium of claim 17, wherein analyzing the text comprises:

associating each of a plurality of entities in the text with one selected from a group consisting of one of the plurality of tags and a noise tag to obtain a set of tags for the text; and

identifying the action using the set of tags for the text and the sentence-level features.

20. The non-transitory computer readable medium of claim 1, wherein at least one of the plurality of features may be generated using a common NLP model, wherein the common NLP model is trained.

21. The non-transitory computer readable medium of claim 20, wherein generating the one of the plurality of features using the common NLP model comprises tagging the training corpus with predefined tags specified in the common NLP model to obtain a set of predefined tags, and generating the one of the plurality of features for one of the plurality of tags using at least one of the set of predefined tags.

22. (canceled)